v0.1.2 - Windows Preview

AgentDock v0.1.2 — Windows Codex Launch Fix

This release fixes a Windows issue where AgentDock could detect the npm codex.cmd shim and incorrectly pass it to the Codex SDK as a native executable.

Fixed

• Codex sessions no longer fail when codex resolves to an npm .cmd shim.
• AgentDock now allows the Codex SDK to use its bundled native Windows runtime by default.
• Added native standalone Codex executable fallback detection.
• Improved custom Codex path validation.
• Improved packaged-runtime diagnostics.
• Added Windows resolver and packaging regression tests.

Users affected by the Windows Codex startup error should install v0.1.2.

AgentDock v0.1.1 — macOS

macOS build of AgentDock v0.1.1.

AgentDock v0.1.1 — Windows

AgentDock v0.1.1 is a major usability and Windows portability improvement.

Windows portability fix

This release fixes a Windows-specific SPAWN EINVAL error that could prevent Claude Code, Codex, or other agents from launching on computers where the CLI was installed as an npm-generated .cmd shim.

AgentDock now safely resolves and launches Windows .cmd, .bat, and .exe commands without unsafe shell-string concatenation.

The fix was manually tested on a separate Windows computer where the previous release failed, and agent launching now works correctly.

New

• Lightweight toggleable workspace file explorer
• Lazy folder expansion
• Automatic file-tree refreshing
• Text and source-code previews
• Image previews
• Optional file-preview panel
• Preview disabled by default
• Persistent preview preference
• File importing into the active workspace
• Rename, replace, skip, and cancel options for import conflicts
• Right-click file and folder context menus
• Open files and folders in Visual Studio Code
• Reveal files in File Explorer
• Copy workspace-relative and full file paths
• Support for attachments beyond images
• Workspace-path fallback for files unsupported by an agent CLI
• Copyable agent-launch diagnostics
• Settings action to reset agent detection and custom executable paths
• Settings action to remove projects whose folders no longer exist

Improved

• Safe Windows .cmd, .bat, and .exe launching
• Claude Code SDK process launching
• Codex SDK process launching
• Antigravity and PTY launching
• Codex model-catalogue startup process
• CLI detection probes
• Process argument, environment, executable, and working-directory validation
• Clearer agent-launch error messages
• Workspace portability
• Correct AgentDock icon in the application, taskbar, installer, and portable executable
• Removed remaining Electron and Vite default icons
• Removed unwanted white and dark boxes around controls
• File explorer layout when preview is disabled
• Browser development-preview initialisation

Technical validation

• Root cause of SPAWN EINVAL identified and fixed
• 59 Windows portability regression tests added
• 551 automated tests passing
• Six pre-existing macOS/POSIX-oriented tests still fail when run on Windows
• TypeScript checks pass
• Production build passes
• Native PTY check passes
• Windows installer packaging passes
• Windows portable packaging passes
• Packaged PTY assets verified
• No development-machine paths found in packaged output
• Correct AgentDock icons verified in final Windows artifacts
• Fixed build manually confirmed on a separate Windows testing PC

UI fixes

• Fixed Claude Code activity details so expanded tool calls show the actual command, file path, and available operation information.
• Reduced excessive spacing between Claude Code tool and activity rows.
• Improved spacing and readability in Antigravity responses and action lines.

Windows downloads

This release includes:

• Windows NSIS installer
• Windows portable executable

AgentDock v0.1.0 — Windows Preview

AgentDock v0.1.0 — Windows Preview

First packaged Windows preview of AgentDock, a desktop interface for running Claude Code, Codex, and Antigravity coding agents on your local projects.

Highlights

• Claude Code, Codex, and Antigravity support — run any of the three coding agents against your own projects from one app.
• Cross-agent continuation and handoffs — hand a conversation off from one agent to another (e.g. Claude → Codex → Antigravity) with full context carried over internally, while the visible chat only ever shows the task you actually typed.
• Persistent projects and conversations — projects, sessions, and message history are saved locally and survive an app restart.
• Automatic Windows CLI discovery — Claude, Codex, and Antigravity executables are found automatically (PATH search, then known install locations), with a manual override available in Settings if needed.
• Installer and portable versions — both a standard Windows installer and a portable, no-install .exe are provided below.

Requirements and limitations

• Windows x64 only for this release — no macOS or Linux build yet.
• Unsigned build — this release is not code-signed, so Windows SmartScreen may show an "unknown publisher" warning when you first run it. Click "More info" → "Run anyway" to proceed.
• Agent CLIs are not bundled — you'll need to separately install whichever of Claude Code, Codex, and/or Antigravity you want to use; AgentDock detects and reports which ones it finds.

Assets

• AgentDock Setup 0.1.0.exe — Windows installer (per-user install, no admin required, Start Menu shortcut, uninstaller included).
• AgentDock 0.1.0.exe — portable build, no installation needed.

AgentDock v0.1.0 — macOS Preview

AgentDock for macOS — First Public Build

The first public macOS build of AgentDock is now available.

AgentDock gives you one desktop workspace for running and managing local AI coding agents without constantly switching between separate terminal windows.

Included

• Native macOS desktop build
• Support for Claude Code, OpenAI Codex CLI, and Antigravity
• Automatically detects installed agent CLIs
• Start new agent sessions from a unified interface
• Continue work between supported agents
• Project and workspace selection
• Chat-style interface for prompts and agent output
• Model, reasoning, effort, and permission controls where supported
• Interactive terminal support
• Local session storage
• Improved cross-platform path and executable handling

Requirements

AgentDock does not include AI subscriptions or API access.

You must install and sign in to the coding agents you want to use separately:

• Claude Code
• OpenAI Codex CLI
• Antigravity

AgentDock runs these tools locally using your existing accounts and subscriptions.

macOS Security Notice

Because this is an early independent release, macOS may warn that the application is from an unidentified developer.

You may need to open:

System Settings → Privacy & Security → Open Anyway

Current Status

This is the first public macOS release and should be treated as an early build.

Some agent-specific features may behave differently depending on the installed CLI version, permissions, model availability, and macOS configuration.

Please report bugs through the GitHub Issues page and include:

• Your macOS version
• Mac model and processor
• Which agent you were using
• Relevant logs or screenshots
• Steps to reproduce the issue

AgentDock is currently a free local desktop tool. More advanced multi-agent functionality is planned for future releases.